I want to make Expert Advisor self-training for MTS in MTS - MQL4 and MetaTrader 4

[Deleted] 2007.10.28 06:04 #101

Vinin:

maveric:

FION:

Victor, have you dealt with the Kohonen maps? I haven't come across any understandable "fish" for multilayer NS. I would like to feel something concrete, even if it does not work well for evaluation. Again - grid training, how many parameters can the computer hold? Although getting into these "rubrics" ..., there is a danger of getting stuck there. Basically, we can optimize the grid by limiting parameters using the same set of indicators.

Somewhere here (in the forum), I posted an advisor does not kohonen but nets layered

I think it will do as a fish.

Could you be more specific? It must have gone by the wayside.

'Neural network training' - all sorts of general thoughts here

"Using artificial intelligence in MTS" - here's an expert attached to my post

I've seen your code, learned something, in particular about normalization of indicator data, but in general I failed to grasp the structure, I got stuck in arrays.

The arrays in arrays are really complicated :( The absence of objects or at least structures in MQL seriously hampers the process in this sense. Either the array is so packed that it cannot be understood without a half-liter, or the packing is clear, but the code is scary :(

If you do not understand arrays, please ask me. I will try to remember and explain.

info about neural network Array references new indicator debugging

Victor Nikolaev 2007.10.28 06:23 #102

maveric:

Vinin:

maveric:

FION:

Victor, have you dealt with the Kohonen maps? I haven't come across any understandable "fish" for multilayer NS. I would like to feel something concrete, even if it does not work well for evaluation. Again - grid training, how many parameters can the computer hold? Although getting into these "rubrics" ..., there is a danger of getting stuck there. Basically, we can optimize the grid by limiting parameters using the same set of indicators.

Somewhere here (in the forum) I posted a tipster Not kohonen, but multilayer nets

I think it will do as a fish.

Can you be more specific? It must have gone by the wayside.

'Neural network training' - all sorts of general thoughts here

"Using Artificial Intelligence in MTS" - there's an expert attached to my post

I've seen your code, learned something, in particular about normalization of indicator data, but in general I failed to grasp the structure, I got stuck in arrays.

The arrays in arrays are really complicated :( The absence of objects or at least structures in MQL seriously hampers the process in this sense. Either the array is so packed that it cannot be understood without a half-liter, or the packing is clear, but the code is scary :(

If you do not understand arrays, please ask me. I will try to remember and explain.

I saw that code, looked through it and remembered it. It was based on the code in C. Translated and adapted. Trade functions were added.

I'm not interested in reverse distribution networks for now. I don't have enough technical capabilities for a normal grid. A small one has no effect.

I want and have started to redesign the Expert Advisor (it takes me much time to optimize it); I will make it self-training. The Kohonen layer is used only for data compression. I have to decide on compression ratio. So far I've taken too much, again due to technical capabilities.

Neuromongers, don't pass by Expert profiling conundrum - Help Please!

Yurixx 2007.10.28 09:06 #103

2 Vinin

Yurixx:

Vinin:

I recommend reading F. Wasserman Neurocomputer Science: Theory and Practice . It is very well written. If you need it, I can send it to you by e-mail. I can read not only this book, but other books as well.

If it's not too much trouble, so am I. My address is in my profile.

I have recently come to the conclusion that without NS my system cannot be taught to trade correctly. As I have seen I am a bad teacher. :-) I have an idea, that it needs proper clustering of data, with which my system works. Well, as far as I understand, they can be clustered using a Kohonen network. But my first attempts to get through all this have not yet led to any results. I know too little about it. I need to read something good that combines both clearly stated ideas and good examples of practical use.

I've read the whole cloth thread on NS, but it's not my level. Need to fill in the blanks immediately.

I repeat my request once again. Unless, of course, your offer was for a select few.

Victor Nikolaev 2007.10.28 09:29 #104

Yurixx:

2 Vinin

Yurixx:

Vinin:

I recommend reading F. Wasserman Neurocomputer Science: Theory and Practice . It is very well written. If you need it, I can send it to you by e-mail. I can read not only this book, but other books as well.

If it's not too much trouble, so am I. My address is in my profile.

I have recently come to the conclusion that without NS my system cannot be taught to trade correctly. As I have seen I am a bad teacher. :-) I have an idea, that it needs proper clustering of data, with which my system works. Well, as far as I understand, they can be clustered using a Kohonen network. But my first attempts to get through all this have not yet led to any results. I know too little about it. I need to read something good that combines both clearly stated ideas and good examples of practical use.

I've read the whole cloth thread on NS, but it's not my level. Need to fill in the blanks immediately.

I repeat my request once again. Unless, of course, your offer was for a select few.

Sent.

Yurixx 2007.10.28 16:27 #105

Vinin:

Yurixx:

I repeat my request once again. Unless, of course, your offer was for a select few.

Sent.

Strange, I haven't received it yet.

Victor Nikolaev 2007.10.28 16:33 #106

Yurixx:

Vinin:

Yurixx:

I repeat my request once again. Unless, of course, your offer was for a select few.

Sent.

Strange, I haven't received it yet.

Send it again?

Yurixx 2007.10.28 21:43 #107

Vinin:

Yurixx:

Vinin:

Yurixx:

I repeat my request once again. Unless, of course, your offer was for a select few.

Sent.

Strange, I haven't received it yet.

Shall I send it again?

Thank you, I got it now.

Leonid Borsky 2007.12.05 23:25 #108

Hello all. I propose to use the so-called "Trend Detector". I wasn't expecting such a good result from this find of mine. Accidentally blinded it - put it in. I insert this part into almost every Expert Advisor and even a losing Expert Advisor yields some profit! It decreases the number of trades against the trend (mostly losing ones) and greatly increases the Profitability parameter of the Expert Advisor, often to at least two! This means that outside the optimization period we are much more likely to make a profit!

Here's the idea: we take BearsPower and BullsPower indicators (bulls power and bears power) and compare them to each other. But just compare them - it's a pain in the neck. To do it programmatically is cumbersome. That's why I put MA on them and compare MA values on zero bar! Just add up these values = Delta. Further everything is simple. If DELTA ..>0 - the trend is up. Otherwise it is going downwards!

We just need to add to the condition to buy if ((Delta>=0) && ... ...

And in the Sell condition - if ((Delta<=0) && ... ...

In external parameters of any Expert Advisor, insert :

//-----------------------------------------------------------
extern int     PeriodPower    =13;
extern int     Period_Bulls   =11;
extern int     Period_Bears   =10;
//-----------------------------------------------------------

You don't have to insert it. But then you have to pick up these parameters and insert numerical values instead of variable names directly into the code. And here is the block itself :

//================ Определитель тренда ============================ 
 double Bears_array[50]; int cx=0; while (cx<51)
 {Bears_array[cx]= iBearsPower(NULL, 0, PeriodPower,PRICE_CLOSE,cx); cx++; }
 ArraySetAsSeries(Bears_array,true);
 double MA_Bears =iMAOnArray(Bears_array,0,Period_Bears,1,MODE_SMMA,0); 
 
 double Bulls_array[50]; int lx=0; while (lx<51)
 {Bulls_array[lx]= iBullsPower(NULL, 0, PeriodPower,PRICE_CLOSE,lx); lx++; } 
 ArraySetAsSeries(Bulls_array,true);
 double MA_Bulls =iMAOnArray(Bulls_array,0,Period_Bulls,1,MODE_SMMA,0); 
 
 double Delta = MA_Bears + MA_Bulls;
 /*Comment ("Delta ", Delta ,"\n") */
//===================================================================

Here is an example of how an EA works with the Trend Detector. We can see that in case of an up trend, the buy positions are opened, and vice versa.

Perhaps someone will have suggestions for improvement and refinement of the design. I would like to know how promising this trend detector will be.

[WARNING CLOSED!] Any newbie Tendential planimetry method Any questions from newcomers

khorosh 2007.12.06 01:14 #109

leonid533.

Please let me know - are these parameters (13,11,10) found for GBPJPY and TF - M30?

Victor Nikolaev 2007.12.06 02:58 #110

leonid553:

Hello all. I propose to use the so-called "Trend Detector". I wasn't expecting such a good result from this find of mine. Accidentally blinded it - put it in. I insert this part into almost every Expert Advisor and even a losing Expert Advisor yields some profit! It decreases the number of trades against the trend (mostly losing ones) and considerably increases the Profitability parameter of the Expert Advisor, often to at least two! This means that outside the optimization period we are much more likely to make a profit!

Here's the idea: we take BearsPower and BullsPower indicators (bulls power and bears power) and compare them to each other. But just compare them - it's a pain in the neck. To do it programmatically is cumbersome. That's why I put MA on them and compare MA values on zero bar! Just add up these values = Delta. Further everything is simple. If DELTA ..>0 - the trend is up. Otherwise it is going downwards!

We just need to add to the condition to buy if ((Delta>=0) && ... ...

And in the Sell condition - if ((Delta<=0) && ... ...

In external parameters of any Expert Advisor, insert :

//-----------------------------------------------------------
extern int     PeriodPower    =13;
extern int     Period_Bulls   =11;
extern int     Period_Bears   =10;
//-----------------------------------------------------------

You don't have to insert it. But then you have to pick up these parameters and insert numerical values instead of variable names directly into the code. And here is the block itself :

//================ Определитель тренда ============================ 
 double Bears_array[50]; int cx=0; while (cx<51)
 {Bears_array[cx]= iBearsPower(NULL, 0, PeriodPower,PRICE_CLOSE,cx); cx++; }
 ArraySetAsSeries(Bears_array,true);
 double MA_Bears =iMAOnArray(Bears_array,0,Period_Bears,1,MODE_SMMA,0); 
 
 double Bulls_array[50]; int lx=0; while (lx<51)
 {Bulls_array[lx]= iBullsPower(NULL, 0, PeriodPower,PRICE_CLOSE,lx); lx++; } 
 ArraySetAsSeries(Bulls_array,true);
 double MA_Bulls =iMAOnArray(Bulls_array,0,Period_Bulls,1,MODE_SMMA,0); 
 
 double Delta = MA_Bears + MA_Bulls;
 /*Comment ("Delta ", Delta ,"\n") */
//===================================================================

Here is an example of how an EA works with the Trend Detector. We can see that in case of an up trend we go to buy, and vice versa.

Perhaps someone will have suggestions for improvement and refinement of the design. I would like to know how promising this trend detector will be.

1. one shows position of High[] relative to the EMA, the second shows position of Low relative to the EMA, so I think we should introduce thresholds (levels of significance). Compare Delta not with zero, but with these thresholds.

2. You have a small error in your code. The variables cx and lx cannot equal 50, the condition must be cx<50 and lx<50. There is an array overrun.

Neural network Question about Operations of Machine learning in trading:

THE IDEA EXCHANGE - page 11